LFHML: Leveraging FTF-XGB and Hybrid Machine Learning for Stroke Prediction from Imbalanced Data

The medical community has widely used machine learning algorithms for rapid prediction and diagnosis of stroke. This study proposes an approach to address the data imbalance problem in stroke disease prediction. More accurate prediction results can be obtained using the "face-to-face" model after data preprocessing and solving imbalance problems. This method only uses features computed directly from the dataset, which is one of its strengths. Then, we use XGBoost for feature selection to reduce the influence of irrelevant variables. Various machine learning methods such as LCE, RF, and XGB are used for feature selection and classifier training, and feature values are gradually reduced to speed up training. The experimental results show that the LCE model achieves the highest accuracy rate of 97.5%, while the XGB and NBC models reach 86.7% and 81.8%, respectively. Studies have shown that this method performs well in predicting stroke disease imbalance in small samples and has reference values for other small sample studies. The proposed method helps improve the accuracy and efficiency of stroke disease prediction and supports the development of mature medical treatment methods.